Incorrect output labels and ECOC matrix dimension in YettonEtAl_MachineLearning detector

[klacourse]

Hi,

I encountered an issue when trying to run the "YettonEtAl_MachineLearning" detector on my dataset.

Steps to reproduce:

1. I ran the detector on my data without any errors.
2. However, the output labels had the dimensions of the components instead of matching the time series (i.e., seconds of the time series).
3. I looked into the ECOC demo and suspect that the data matrices need to be transposed, including the classes matrix.
4. After transposing, I now receive an error when running the "YettonEtAl_MachineLearning" detector.

Issue:

• It seems the error is linked to the ECOC matrix dimensions. The ECOC matrix in the repository is currently 3x3, but based on the demo, it seems like the matrix should be n_classes x n_components. For my case, this would mean the ECOC matrix should be 3x33 instead of 3x3.
• The current ECOC matrix is being loaded directly from the GitHub repository.

Question:

Am I missing something in the data preparation or processing? Should the ECOC matrix indeed be resized to 3x33, and if so, could the preloaded matrix in the repository be incorrect?

Any guidance on how to resolve this would be greatly appreciated!

Thanks in advance!

[bdyetton]

Hi Karine, I think your best bet here is to use the HatzilabrouEtAl algorithm. It is:

• Faster
• Less fine tuned for the specific datasets we used
• Seemed to work for others sufficiently well Note also that this is very old code now, and the ML methods i used are rather outdated.

[klacourse]

Hi Ben,

Thanks for your fast answer. I successfully ran the HatzilabrouEtAl algorithm on my own data.

Best regards, Karine